A data centre is a physical facility that houses computing, storage and networking equipment. Every cloud file, streamed video, website request and AI output ultimately depends on equipment running somewhere—not an immaterial “cloud.”
This beginner’s guide explains where the energy goes, why location and timing matter, and how to read environmental claims without reducing a complex system to one attractive number.
Where does data-centre electricity go?
- IT equipment: servers, storage and network devices doing the computational work.
- Cooling: systems that remove heat and maintain safe operating conditions.
- Power conversion and distribution: transformers, uninterruptible power supplies and other electrical equipment.
- Supporting systems: lighting, controls, security and facility operations.
Demand is growing quickly. The International Energy Agency reported that global data-centre electricity demand grew by 17% in 2025, with AI-focused facilities growing faster. Its projections remain scenarios rather than certainties, but they make efficiency and clean power increasingly consequential. See the IEA’s latest overview.
The difference between power and energy
Power, measured in watts, is the rate at which electricity is used at a moment. Energy, often measured in kilowatt-hours, is power used over time. A 1-kilowatt load running for two hours uses 2 kilowatt-hours.
Capacity announcements are often expressed in megawatts, while environmental reporting may use annual megawatt-hours or terawatt-hours. Confusing the two can produce wildly inaccurate comparisons.
Four metrics worth understanding
| Metric | What it describes | What it does not prove |
|---|---|---|
| PUE | Total facility energy divided by IT-equipment energy | That the electricity is low-carbon |
| WUE | Water use relative to IT energy or another defined output | The local water impact without context |
| CUE | Operational carbon emissions relative to IT energy | Full lifecycle emissions |
| Utilisation | How much available computing capacity is doing work | That the work itself is valuable or efficient |
A PUE close to 1 means less facility overhead relative to IT energy. It does not mean the servers are efficient, fully used or powered by low-carbon electricity. Compare like with like: annual averages and short-term readings, climates, facility types and measurement boundaries can differ.
Why the electricity source matters
The same workload can produce different operational emissions depending on when and where it runs. An annual renewable-energy contract is useful information, but it is not identical to matching consumption with carbon-free generation every hour.
Good reporting distinguishes energy efficiency from electricity procurement and states whether emissions figures are location-based, market-based or calculated another way.
Why water use varies
Some cooling systems consume water directly; electricity generation can also have an indirect water footprint. Climate, cooling technology, water source and local scarcity all affect the significance of a headline volume. A litre used in a water-stressed region is not equivalent in impact to a litre from a plentiful reclaimed source.
UN Environment Programme guidance emphasises that water impacts vary greatly with design, cooling and location.
Embodied impacts and electronic waste
Operational electricity is only part of the lifecycle. Servers and buildings require raw materials, manufacturing and transport; replaced equipment becomes an end-of-life responsibility. Extending useful hardware life and increasing productive utilisation can reduce embodied emissions per unit of computing, provided obsolete equipment is handled securely and responsibly.
How operators reduce impact
- Improve server utilisation and retire idle equipment.
- Use efficient hardware appropriate to the workload.
- Reduce cooling and power-distribution losses.
- Locate and schedule flexible workloads with grid and water conditions in mind.
- Contract for additional clean electricity and report matching limitations.
- Design equipment for longer use, repair, reuse and responsible recycling.
- Measure energy, emissions and water with transparent boundaries.
What software teams can influence
Developers may not select the cooling system, but they influence how much infrastructure an application demands. Avoid unnecessary computation and data transfer, right-size services, shut down unused environments, choose suitable regions and architectures, and schedule delay-tolerant jobs carefully.
Our guide to energy-efficient code turns those principles into software decisions.
How to evaluate a provider’s sustainability claim
- Find the reporting period and geographic boundary.
- Separate total facility energy from IT energy.
- Check whether PUE is annual, measured and independently assured.
- Ask how electricity emissions are calculated.
- Look for absolute energy and emissions, not intensity alone.
- Check water source, local scarcity and cooling method.
- Look for hardware lifecycle and e-waste information.
- Note missing data and changes in methodology.
The short version
A more sustainable data centre needs efficient facilities, efficient and well-used IT equipment, cleaner electricity, responsible water management and longer-lived hardware. No single metric captures all five. Start with transparent boundaries, compare consistent measurements and remain cautious about precise claims that cannot be audited.
For the broader lifecycle, read what sustainable computing means. For the fastest-growing new demand, see is AI sustainable?